projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
ipv4: Namespaceify tcp syncookies sysctl knob
[deliverable/linux.git] / net / ipv4 / tcp_ipv4.c
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Implementation of the Transmission Control Protocol(TCP).
7 *
8 * IPv4 specific functions
9 *
10 *
11 * code split from:
12 * linux/ipv4/tcp.c
13 * linux/ipv4/tcp_input.c
14 * linux/ipv4/tcp_output.c
15 *
16 * See tcp.c for author information
17 *
18 * This program is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU General Public License
20 * as published by the Free Software Foundation; either version
21 * 2 of the License, or (at your option) any later version.
22 */
23
24 /*
25 * Changes:
26 * David S. Miller : New socket lookup architecture.
27 * This code is dedicated to John Dyson.
28 * David S. Miller : Change semantics of established hash,
29 * half is devoted to TIME_WAIT sockets
30 * and the rest go in the other half.
31 * Andi Kleen : Add support for syncookies and fixed
32 * some bugs: ip options weren't passed to
33 * the TCP layer, missed a check for an
34 * ACK bit.
35 * Andi Kleen : Implemented fast path mtu discovery.
36 * Fixed many serious bugs in the
37 * request_sock handling and moved
38 * most of it into the af independent code.
39 * Added tail drop and some other bugfixes.
40 * Added new listen semantics.
41 * Mike McLagan : Routing by source
42 * Juan Jose Ciarlante: ip_dynaddr bits
43 * Andi Kleen: various fixes.
44 * Vitaly E. Lavrov : Transparent proxy revived after year
45 * coma.
46 * Andi Kleen : Fix new listen.
47 * Andi Kleen : Fix accept error reporting.
48 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
49 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
50 * a single port at the same time.
51 */
52
53 #define pr_fmt(fmt) "TCP: " fmt
54
55 #include <linux/bottom_half.h>
56 #include <linux/types.h>
57 #include <linux/fcntl.h>
58 #include <linux/module.h>
59 #include <linux/random.h>
60 #include <linux/cache.h>
61 #include <linux/jhash.h>
62 #include <linux/init.h>
63 #include <linux/times.h>
64 #include <linux/slab.h>
65
66 #include <net/net_namespace.h>
67 #include <net/icmp.h>
68 #include <net/inet_hashtables.h>
69 #include <net/tcp.h>
70 #include <net/transp_v6.h>
71 #include <net/ipv6.h>
72 #include <net/inet_common.h>
73 #include <net/timewait_sock.h>
74 #include <net/xfrm.h>
75 #include <net/secure_seq.h>
76 #include <net/busy_poll.h>
77
78 #include <linux/inet.h>
79 #include <linux/ipv6.h>
80 #include <linux/stddef.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
83
84 #include <linux/crypto.h>
85 #include <linux/scatterlist.h>
86
87 int sysctl_tcp_tw_reuse __read_mostly;
88 int sysctl_tcp_low_latency __read_mostly;
89 EXPORT_SYMBOL(sysctl_tcp_low_latency);
90
91 #ifdef CONFIG_TCP_MD5SIG
92 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
93 __be32 daddr, __be32 saddr, const struct tcphdr *th);
94 #endif
95
96 struct inet_hashinfo tcp_hashinfo;
97 EXPORT_SYMBOL(tcp_hashinfo);
98
99 static __u32 tcp_v4_init_sequence(const struct sk_buff *skb)
100 {
101 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
102 ip_hdr(skb)->saddr,
103 tcp_hdr(skb)->dest,
104 tcp_hdr(skb)->source);
105 }
106
107 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
108 {
109 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
110 struct tcp_sock *tp = tcp_sk(sk);
111
112 /* With PAWS, it is safe from the viewpoint
113 of data integrity. Even without PAWS it is safe provided sequence
114 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
115
116 Actually, the idea is close to VJ's one, only timestamp cache is
117 held not per host, but per port pair and TW bucket is used as state
118 holder.
119
120 If TW bucket has been already destroyed we fall back to VJ's scheme
121 and use initial timestamp retrieved from peer table.
122 */
123 if (tcptw->tw_ts_recent_stamp &&
124 (!twp || (sysctl_tcp_tw_reuse &&
125 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
126 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
127 if (tp->write_seq == 0)
128 tp->write_seq = 1;
129 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
130 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
131 sock_hold(sktw);
132 return 1;
133 }
134
135 return 0;
136 }
137 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
138
139 /* This will initiate an outgoing connection. */
140 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
141 {
142 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
143 struct inet_sock *inet = inet_sk(sk);
144 struct tcp_sock *tp = tcp_sk(sk);
145 __be16 orig_sport, orig_dport;
146 __be32 daddr, nexthop;
147 struct flowi4 *fl4;
148 struct rtable *rt;
149 int err;
150 struct ip_options_rcu *inet_opt;
151
152 if (addr_len < sizeof(struct sockaddr_in))
153 return -EINVAL;
154
155 if (usin->sin_family != AF_INET)
156 return -EAFNOSUPPORT;
157
158 nexthop = daddr = usin->sin_addr.s_addr;
159 inet_opt = rcu_dereference_protected(inet->inet_opt,
160 sock_owned_by_user(sk));
161 if (inet_opt && inet_opt->opt.srr) {
162 if (!daddr)
163 return -EINVAL;
164 nexthop = inet_opt->opt.faddr;
165 }
166
167 orig_sport = inet->inet_sport;
168 orig_dport = usin->sin_port;
169 fl4 = &inet->cork.fl.u.ip4;
170 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
171 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
172 IPPROTO_TCP,
173 orig_sport, orig_dport, sk);
174 if (IS_ERR(rt)) {
175 err = PTR_ERR(rt);
176 if (err == -ENETUNREACH)
177 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
178 return err;
179 }
180
181 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
182 ip_rt_put(rt);
183 return -ENETUNREACH;
184 }
185
186 if (!inet_opt || !inet_opt->opt.srr)
187 daddr = fl4->daddr;
188
189 if (!inet->inet_saddr)
190 inet->inet_saddr = fl4->saddr;
191 sk_rcv_saddr_set(sk, inet->inet_saddr);
192
193 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
194 /* Reset inherited state */
195 tp->rx_opt.ts_recent = 0;
196 tp->rx_opt.ts_recent_stamp = 0;
197 if (likely(!tp->repair))
198 tp->write_seq = 0;
199 }
200
201 if (tcp_death_row.sysctl_tw_recycle &&
202 !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr)
203 tcp_fetch_timewait_stamp(sk, &rt->dst);
204
205 inet->inet_dport = usin->sin_port;
206 sk_daddr_set(sk, daddr);
207
208 inet_csk(sk)->icsk_ext_hdr_len = 0;
209 if (inet_opt)
210 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
211
212 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
213
214 /* Socket identity is still unknown (sport may be zero).
215 * However we set state to SYN-SENT and not releasing socket
216 * lock select source port, enter ourselves into the hash tables and
217 * complete initialization after this.
218 */
219 tcp_set_state(sk, TCP_SYN_SENT);
220 err = inet_hash_connect(&tcp_death_row, sk);
221 if (err)
222 goto failure;
223
224 sk_set_txhash(sk);
225
226 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
227 inet->inet_sport, inet->inet_dport, sk);
228 if (IS_ERR(rt)) {
229 err = PTR_ERR(rt);
230 rt = NULL;
231 goto failure;
232 }
233 /* OK, now commit destination to socket. */
234 sk->sk_gso_type = SKB_GSO_TCPV4;
235 sk_setup_caps(sk, &rt->dst);
236
237 if (!tp->write_seq && likely(!tp->repair))
238 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
239 inet->inet_daddr,
240 inet->inet_sport,
241 usin->sin_port);
242
243 inet->inet_id = tp->write_seq ^ jiffies;
244
245 err = tcp_connect(sk);
246
247 rt = NULL;
248 if (err)
249 goto failure;
250
251 return 0;
252
253 failure:
254 /*
255 * This unhashes the socket and releases the local port,
256 * if necessary.
257 */
258 tcp_set_state(sk, TCP_CLOSE);
259 ip_rt_put(rt);
260 sk->sk_route_caps = 0;
261 inet->inet_dport = 0;
262 return err;
263 }
264 EXPORT_SYMBOL(tcp_v4_connect);
265
266 /*
267 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
268 * It can be called through tcp_release_cb() if socket was owned by user
269 * at the time tcp_v4_err() was called to handle ICMP message.
270 */
271 void tcp_v4_mtu_reduced(struct sock *sk)
272 {
273 struct dst_entry *dst;
274 struct inet_sock *inet = inet_sk(sk);
275 u32 mtu = tcp_sk(sk)->mtu_info;
276
277 dst = inet_csk_update_pmtu(sk, mtu);
278 if (!dst)
279 return;
280
281 /* Something is about to be wrong... Remember soft error
282 * for the case, if this connection will not able to recover.
283 */
284 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
285 sk->sk_err_soft = EMSGSIZE;
286
287 mtu = dst_mtu(dst);
288
289 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
290 ip_sk_accept_pmtu(sk) &&
291 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
292 tcp_sync_mss(sk, mtu);
293
294 /* Resend the TCP packet because it's
295 * clear that the old packet has been
296 * dropped. This is the new "fast" path mtu
297 * discovery.
298 */
299 tcp_simple_retransmit(sk);
300 } /* else let the usual retransmit timer handle it */
301 }
302 EXPORT_SYMBOL(tcp_v4_mtu_reduced);
303
304 static void do_redirect(struct sk_buff *skb, struct sock *sk)
305 {
306 struct dst_entry *dst = __sk_dst_check(sk, 0);
307
308 if (dst)
309 dst->ops->redirect(dst, sk, skb);
310 }
311
312
313 /* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
314 void tcp_req_err(struct sock *sk, u32 seq)
315 {
316 struct request_sock *req = inet_reqsk(sk);
317 struct net *net = sock_net(sk);
318
319 /* ICMPs are not backlogged, hence we cannot get
320 * an established socket here.
321 */
322 WARN_ON(req->sk);
323
324 if (seq != tcp_rsk(req)->snt_isn) {
325 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
326 } else {
327 /*
328 * Still in SYN_RECV, just remove it silently.
329 * There is no good way to pass the error to the newly
330 * created socket, and POSIX does not want network
331 * errors returned from accept().
332 */
333 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
334 NET_INC_STATS_BH(net, LINUX_MIB_LISTENDROPS);
335 }
336 reqsk_put(req);
337 }
338 EXPORT_SYMBOL(tcp_req_err);
339
340 /*
341 * This routine is called by the ICMP module when it gets some
342 * sort of error condition. If err < 0 then the socket should
343 * be closed and the error returned to the user. If err > 0
344 * it's just the icmp type << 8 | icmp code. After adjustment
345 * header points to the first 8 bytes of the tcp header. We need
346 * to find the appropriate port.
347 *
348 * The locking strategy used here is very "optimistic". When
349 * someone else accesses the socket the ICMP is just dropped
350 * and for some paths there is no check at all.
351 * A more general error queue to queue errors for later handling
352 * is probably better.
353 *
354 */
355
356 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
357 {
358 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
359 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
360 struct inet_connection_sock *icsk;
361 struct tcp_sock *tp;
362 struct inet_sock *inet;
363 const int type = icmp_hdr(icmp_skb)->type;
364 const int code = icmp_hdr(icmp_skb)->code;
365 struct sock *sk;
366 struct sk_buff *skb;
367 struct request_sock *fastopen;
368 __u32 seq, snd_una;
369 __u32 remaining;
370 int err;
371 struct net *net = dev_net(icmp_skb->dev);
372
373 sk = __inet_lookup_established(net, &tcp_hashinfo, iph->daddr,
374 th->dest, iph->saddr, ntohs(th->source),
375 inet_iif(icmp_skb));
376 if (!sk) {
377 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
378 return;
379 }
380 if (sk->sk_state == TCP_TIME_WAIT) {
381 inet_twsk_put(inet_twsk(sk));
382 return;
383 }
384 seq = ntohl(th->seq);
385 if (sk->sk_state == TCP_NEW_SYN_RECV)
386 return tcp_req_err(sk, seq);
387
388 bh_lock_sock(sk);
389 /* If too many ICMPs get dropped on busy
390 * servers this needs to be solved differently.
391 * We do take care of PMTU discovery (RFC1191) special case :
392 * we can receive locally generated ICMP messages while socket is held.
393 */
394 if (sock_owned_by_user(sk)) {
395 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
396 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
397 }
398 if (sk->sk_state == TCP_CLOSE)
399 goto out;
400
401 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
402 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
403 goto out;
404 }
405
406 icsk = inet_csk(sk);
407 tp = tcp_sk(sk);
408 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
409 fastopen = tp->fastopen_rsk;
410 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
411 if (sk->sk_state != TCP_LISTEN &&
412 !between(seq, snd_una, tp->snd_nxt)) {
413 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
414 goto out;
415 }
416
417 switch (type) {
418 case ICMP_REDIRECT:
419 do_redirect(icmp_skb, sk);
420 goto out;
421 case ICMP_SOURCE_QUENCH:
422 /* Just silently ignore these. */
423 goto out;
424 case ICMP_PARAMETERPROB:
425 err = EPROTO;
426 break;
427 case ICMP_DEST_UNREACH:
428 if (code > NR_ICMP_UNREACH)
429 goto out;
430
431 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
432 /* We are not interested in TCP_LISTEN and open_requests
433 * (SYN-ACKs send out by Linux are always <576bytes so
434 * they should go through unfragmented).
435 */
436 if (sk->sk_state == TCP_LISTEN)
437 goto out;
438
439 tp->mtu_info = info;
440 if (!sock_owned_by_user(sk)) {
441 tcp_v4_mtu_reduced(sk);
442 } else {
443 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &tp->tsq_flags))
444 sock_hold(sk);
445 }
446 goto out;
447 }
448
449 err = icmp_err_convert[code].errno;
450 /* check if icmp_skb allows revert of backoff
451 * (see draft-zimmermann-tcp-lcd) */
452 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
453 break;
454 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
455 !icsk->icsk_backoff || fastopen)
456 break;
457
458 if (sock_owned_by_user(sk))
459 break;
460
461 icsk->icsk_backoff--;
462 icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) :
463 TCP_TIMEOUT_INIT;
464 icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
465
466 skb = tcp_write_queue_head(sk);
467 BUG_ON(!skb);
468
469 remaining = icsk->icsk_rto -
470 min(icsk->icsk_rto,
471 tcp_time_stamp - tcp_skb_timestamp(skb));
472
473 if (remaining) {
474 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
475 remaining, TCP_RTO_MAX);
476 } else {
477 /* RTO revert clocked out retransmission.
478 * Will retransmit now */
479 tcp_retransmit_timer(sk);
480 }
481
482 break;
483 case ICMP_TIME_EXCEEDED:
484 err = EHOSTUNREACH;
485 break;
486 default:
487 goto out;
488 }
489
490 switch (sk->sk_state) {
491 case TCP_SYN_SENT:
492 case TCP_SYN_RECV:
493 /* Only in fast or simultaneous open. If a fast open socket is
494 * is already accepted it is treated as a connected one below.
495 */
496 if (fastopen && !fastopen->sk)
497 break;
498
499 if (!sock_owned_by_user(sk)) {
500 sk->sk_err = err;
501
502 sk->sk_error_report(sk);
503
504 tcp_done(sk);
505 } else {
506 sk->sk_err_soft = err;
507 }
508 goto out;
509 }
510
511 /* If we've already connected we will keep trying
512 * until we time out, or the user gives up.
513 *
514 * rfc1122 4.2.3.9 allows to consider as hard errors
515 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
516 * but it is obsoleted by pmtu discovery).
517 *
518 * Note, that in modern internet, where routing is unreliable
519 * and in each dark corner broken firewalls sit, sending random
520 * errors ordered by their masters even this two messages finally lose
521 * their original sense (even Linux sends invalid PORT_UNREACHs)
522 *
523 * Now we are in compliance with RFCs.
524 * --ANK (980905)
525 */
526
527 inet = inet_sk(sk);
528 if (!sock_owned_by_user(sk) && inet->recverr) {
529 sk->sk_err = err;
530 sk->sk_error_report(sk);
531 } else { /* Only an error on timeout */
532 sk->sk_err_soft = err;
533 }
534
535 out:
536 bh_unlock_sock(sk);
537 sock_put(sk);
538 }
539
540 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
541 {
542 struct tcphdr *th = tcp_hdr(skb);
543
544 if (skb->ip_summed == CHECKSUM_PARTIAL) {
545 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
546 skb->csum_start = skb_transport_header(skb) - skb->head;
547 skb->csum_offset = offsetof(struct tcphdr, check);
548 } else {
549 th->check = tcp_v4_check(skb->len, saddr, daddr,
550 csum_partial(th,
551 th->doff << 2,
552 skb->csum));
553 }
554 }
555
556 /* This routine computes an IPv4 TCP checksum. */
557 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
558 {
559 const struct inet_sock *inet = inet_sk(sk);
560
561 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
562 }
563 EXPORT_SYMBOL(tcp_v4_send_check);
564
565 /*
566 * This routine will send an RST to the other tcp.
567 *
568 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
569 * for reset.
570 * Answer: if a packet caused RST, it is not for a socket
571 * existing in our system, if it is matched to a socket,
572 * it is just duplicate segment or bug in other side's TCP.
573 * So that we build reply only basing on parameters
574 * arrived with segment.
575 * Exception: precedence violation. We do not implement it in any case.
576 */
577
578 static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
579 {
580 const struct tcphdr *th = tcp_hdr(skb);
581 struct {
582 struct tcphdr th;
583 #ifdef CONFIG_TCP_MD5SIG
584 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
585 #endif
586 } rep;
587 struct ip_reply_arg arg;
588 #ifdef CONFIG_TCP_MD5SIG
589 struct tcp_md5sig_key *key = NULL;
590 const __u8 *hash_location = NULL;
591 unsigned char newhash[16];
592 int genhash;
593 struct sock *sk1 = NULL;
594 #endif
595 struct net *net;
596
597 /* Never send a reset in response to a reset. */
598 if (th->rst)
599 return;
600
601 /* If sk not NULL, it means we did a successful lookup and incoming
602 * route had to be correct. prequeue might have dropped our dst.
603 */
604 if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
605 return;
606
607 /* Swap the send and the receive. */
608 memset(&rep, 0, sizeof(rep));
609 rep.th.dest = th->source;
610 rep.th.source = th->dest;
611 rep.th.doff = sizeof(struct tcphdr) / 4;
612 rep.th.rst = 1;
613
614 if (th->ack) {
615 rep.th.seq = th->ack_seq;
616 } else {
617 rep.th.ack = 1;
618 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
619 skb->len - (th->doff << 2));
620 }
621
622 memset(&arg, 0, sizeof(arg));
623 arg.iov[0].iov_base = (unsigned char *)&rep;
624 arg.iov[0].iov_len = sizeof(rep.th);
625
626 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
627 #ifdef CONFIG_TCP_MD5SIG
628 hash_location = tcp_parse_md5sig_option(th);
629 if (sk && sk_fullsock(sk)) {
630 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
631 &ip_hdr(skb)->saddr, AF_INET);
632 } else if (hash_location) {
633 /*
634 * active side is lost. Try to find listening socket through
635 * source port, and then find md5 key through listening socket.
636 * we are not loose security here:
637 * Incoming packet is checked with md5 hash with finding key,
638 * no RST generated if md5 hash doesn't match.
639 */
640 sk1 = __inet_lookup_listener(net,
641 &tcp_hashinfo, ip_hdr(skb)->saddr,
642 th->source, ip_hdr(skb)->daddr,
643 ntohs(th->source), inet_iif(skb));
644 /* don't send rst if it can't find key */
645 if (!sk1)
646 return;
647 rcu_read_lock();
648 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
649 &ip_hdr(skb)->saddr, AF_INET);
650 if (!key)
651 goto release_sk1;
652
653 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
654 if (genhash || memcmp(hash_location, newhash, 16) != 0)
655 goto release_sk1;
656 }
657
658 if (key) {
659 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
660 (TCPOPT_NOP << 16) |
661 (TCPOPT_MD5SIG << 8) |
662 TCPOLEN_MD5SIG);
663 /* Update length and the length the header thinks exists */
664 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
665 rep.th.doff = arg.iov[0].iov_len / 4;
666
667 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
668 key, ip_hdr(skb)->saddr,
669 ip_hdr(skb)->daddr, &rep.th);
670 }
671 #endif
672 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
673 ip_hdr(skb)->saddr, /* XXX */
674 arg.iov[0].iov_len, IPPROTO_TCP, 0);
675 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
676 arg.flags = (sk && inet_sk_transparent(sk)) ? IP_REPLY_ARG_NOSRCCHECK : 0;
677
678 /* When socket is gone, all binding information is lost.
679 * routing might fail in this case. No choice here, if we choose to force
680 * input interface, we will misroute in case of asymmetric route.
681 */
682 if (sk)
683 arg.bound_dev_if = sk->sk_bound_dev_if;
684
685 BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) !=
686 offsetof(struct inet_timewait_sock, tw_bound_dev_if));
687
688 arg.tos = ip_hdr(skb)->tos;
689 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
690 skb, &TCP_SKB_CB(skb)->header.h4.opt,
691 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
692 &arg, arg.iov[0].iov_len);
693
694 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
695 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
696
697 #ifdef CONFIG_TCP_MD5SIG
698 release_sk1:
699 if (sk1) {
700 rcu_read_unlock();
701 sock_put(sk1);
702 }
703 #endif
704 }
705
706 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
707 outside socket context is ugly, certainly. What can I do?
708 */
709
710 static void tcp_v4_send_ack(struct net *net,
711 struct sk_buff *skb, u32 seq, u32 ack,
712 u32 win, u32 tsval, u32 tsecr, int oif,
713 struct tcp_md5sig_key *key,
714 int reply_flags, u8 tos)
715 {
716 const struct tcphdr *th = tcp_hdr(skb);
717 struct {
718 struct tcphdr th;
719 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
720 #ifdef CONFIG_TCP_MD5SIG
721 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
722 #endif
723 ];
724 } rep;
725 struct ip_reply_arg arg;
726
727 memset(&rep.th, 0, sizeof(struct tcphdr));
728 memset(&arg, 0, sizeof(arg));
729
730 arg.iov[0].iov_base = (unsigned char *)&rep;
731 arg.iov[0].iov_len = sizeof(rep.th);
732 if (tsecr) {
733 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
734 (TCPOPT_TIMESTAMP << 8) |
735 TCPOLEN_TIMESTAMP);
736 rep.opt[1] = htonl(tsval);
737 rep.opt[2] = htonl(tsecr);
738 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
739 }
740
741 /* Swap the send and the receive. */
742 rep.th.dest = th->source;
743 rep.th.source = th->dest;
744 rep.th.doff = arg.iov[0].iov_len / 4;
745 rep.th.seq = htonl(seq);
746 rep.th.ack_seq = htonl(ack);
747 rep.th.ack = 1;
748 rep.th.window = htons(win);
749
750 #ifdef CONFIG_TCP_MD5SIG
751 if (key) {
752 int offset = (tsecr) ? 3 : 0;
753
754 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
755 (TCPOPT_NOP << 16) |
756 (TCPOPT_MD5SIG << 8) |
757 TCPOLEN_MD5SIG);
758 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
759 rep.th.doff = arg.iov[0].iov_len/4;
760
761 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
762 key, ip_hdr(skb)->saddr,
763 ip_hdr(skb)->daddr, &rep.th);
764 }
765 #endif
766 arg.flags = reply_flags;
767 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
768 ip_hdr(skb)->saddr, /* XXX */
769 arg.iov[0].iov_len, IPPROTO_TCP, 0);
770 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
771 if (oif)
772 arg.bound_dev_if = oif;
773 arg.tos = tos;
774 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
775 skb, &TCP_SKB_CB(skb)->header.h4.opt,
776 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
777 &arg, arg.iov[0].iov_len);
778
779 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
780 }
781
782 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
783 {
784 struct inet_timewait_sock *tw = inet_twsk(sk);
785 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
786
787 tcp_v4_send_ack(sock_net(sk), skb,
788 tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
789 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
790 tcp_time_stamp + tcptw->tw_ts_offset,
791 tcptw->tw_ts_recent,
792 tw->tw_bound_dev_if,
793 tcp_twsk_md5_key(tcptw),
794 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
795 tw->tw_tos
796 );
797
798 inet_twsk_put(tw);
799 }
800
801 static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
802 struct request_sock *req)
803 {
804 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
805 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
806 */
807 u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 :
808 tcp_sk(sk)->snd_nxt;
809
810 tcp_v4_send_ack(sock_net(sk), skb, seq,
811 tcp_rsk(req)->rcv_nxt, req->rsk_rcv_wnd,
812 tcp_time_stamp,
813 req->ts_recent,
814 0,
815 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
816 AF_INET),
817 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
818 ip_hdr(skb)->tos);
819 }
820
821 /*
822 * Send a SYN-ACK after having received a SYN.
823 * This still operates on a request_sock only, not on a big
824 * socket.
825 */
826 static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
827 struct flowi *fl,
828 struct request_sock *req,
829 struct tcp_fastopen_cookie *foc,
830 bool attach_req)
831 {
832 const struct inet_request_sock *ireq = inet_rsk(req);
833 struct flowi4 fl4;
834 int err = -1;
835 struct sk_buff *skb;
836
837 /* First, grab a route. */
838 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
839 return -1;
840
841 skb = tcp_make_synack(sk, dst, req, foc, attach_req);
842
843 if (skb) {
844 __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
845
846 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
847 ireq->ir_rmt_addr,
848 ireq->opt);
849 err = net_xmit_eval(err);
850 }
851
852 return err;
853 }
854
855 /*
856 * IPv4 request_sock destructor.
857 */
858 static void tcp_v4_reqsk_destructor(struct request_sock *req)
859 {
860 kfree(inet_rsk(req)->opt);
861 }
862
863 #ifdef CONFIG_TCP_MD5SIG
864 /*
865 * RFC2385 MD5 checksumming requires a mapping of
866 * IP address->MD5 Key.
867 * We need to maintain these in the sk structure.
868 */
869
870 /* Find the Key structure for an address. */
871 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
872 const union tcp_md5_addr *addr,
873 int family)
874 {
875 const struct tcp_sock *tp = tcp_sk(sk);
876 struct tcp_md5sig_key *key;
877 unsigned int size = sizeof(struct in_addr);
878 const struct tcp_md5sig_info *md5sig;
879
880 /* caller either holds rcu_read_lock() or socket lock */
881 md5sig = rcu_dereference_check(tp->md5sig_info,
882 sock_owned_by_user(sk) ||
883 lockdep_is_held((spinlock_t *)&sk->sk_lock.slock));
884 if (!md5sig)
885 return NULL;
886 #if IS_ENABLED(CONFIG_IPV6)
887 if (family == AF_INET6)
888 size = sizeof(struct in6_addr);
889 #endif
890 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
891 if (key->family != family)
892 continue;
893 if (!memcmp(&key->addr, addr, size))
894 return key;
895 }
896 return NULL;
897 }
898 EXPORT_SYMBOL(tcp_md5_do_lookup);
899
900 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
901 const struct sock *addr_sk)
902 {
903 const union tcp_md5_addr *addr;
904
905 addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
906 return tcp_md5_do_lookup(sk, addr, AF_INET);
907 }
908 EXPORT_SYMBOL(tcp_v4_md5_lookup);
909
910 /* This can be called on a newly created socket, from other files */
911 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
912 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
913 {
914 /* Add Key to the list */
915 struct tcp_md5sig_key *key;
916 struct tcp_sock *tp = tcp_sk(sk);
917 struct tcp_md5sig_info *md5sig;
918
919 key = tcp_md5_do_lookup(sk, addr, family);
920 if (key) {
921 /* Pre-existing entry - just update that one. */
922 memcpy(key->key, newkey, newkeylen);
923 key->keylen = newkeylen;
924 return 0;
925 }
926
927 md5sig = rcu_dereference_protected(tp->md5sig_info,
928 sock_owned_by_user(sk) ||
929 lockdep_is_held(&sk->sk_lock.slock));
930 if (!md5sig) {
931 md5sig = kmalloc(sizeof(*md5sig), gfp);
932 if (!md5sig)
933 return -ENOMEM;
934
935 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
936 INIT_HLIST_HEAD(&md5sig->head);
937 rcu_assign_pointer(tp->md5sig_info, md5sig);
938 }
939
940 key = sock_kmalloc(sk, sizeof(*key), gfp);
941 if (!key)
942 return -ENOMEM;
943 if (!tcp_alloc_md5sig_pool()) {
944 sock_kfree_s(sk, key, sizeof(*key));
945 return -ENOMEM;
946 }
947
948 memcpy(key->key, newkey, newkeylen);
949 key->keylen = newkeylen;
950 key->family = family;
951 memcpy(&key->addr, addr,
952 (family == AF_INET6) ? sizeof(struct in6_addr) :
953 sizeof(struct in_addr));
954 hlist_add_head_rcu(&key->node, &md5sig->head);
955 return 0;
956 }
957 EXPORT_SYMBOL(tcp_md5_do_add);
958
959 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
960 {
961 struct tcp_md5sig_key *key;
962
963 key = tcp_md5_do_lookup(sk, addr, family);
964 if (!key)
965 return -ENOENT;
966 hlist_del_rcu(&key->node);
967 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
968 kfree_rcu(key, rcu);
969 return 0;
970 }
971 EXPORT_SYMBOL(tcp_md5_do_del);
972
973 static void tcp_clear_md5_list(struct sock *sk)
974 {
975 struct tcp_sock *tp = tcp_sk(sk);
976 struct tcp_md5sig_key *key;
977 struct hlist_node *n;
978 struct tcp_md5sig_info *md5sig;
979
980 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
981
982 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
983 hlist_del_rcu(&key->node);
984 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
985 kfree_rcu(key, rcu);
986 }
987 }
988
989 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
990 int optlen)
991 {
992 struct tcp_md5sig cmd;
993 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
994
995 if (optlen < sizeof(cmd))
996 return -EINVAL;
997
998 if (copy_from_user(&cmd, optval, sizeof(cmd)))
999 return -EFAULT;
1000
1001 if (sin->sin_family != AF_INET)
1002 return -EINVAL;
1003
1004 if (!cmd.tcpm_keylen)
1005 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1006 AF_INET);
1007
1008 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1009 return -EINVAL;
1010
1011 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1012 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1013 GFP_KERNEL);
1014 }
1015
1016 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1017 __be32 daddr, __be32 saddr, int nbytes)
1018 {
1019 struct tcp4_pseudohdr *bp;
1020 struct scatterlist sg;
1021
1022 bp = &hp->md5_blk.ip4;
1023
1024 /*
1025 * 1. the TCP pseudo-header (in the order: source IP address,
1026 * destination IP address, zero-padded protocol number, and
1027 * segment length)
1028 */
1029 bp->saddr = saddr;
1030 bp->daddr = daddr;
1031 bp->pad = 0;
1032 bp->protocol = IPPROTO_TCP;
1033 bp->len = cpu_to_be16(nbytes);
1034
1035 sg_init_one(&sg, bp, sizeof(*bp));
1036 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1037 }
1038
1039 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1040 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1041 {
1042 struct tcp_md5sig_pool *hp;
1043 struct hash_desc *desc;
1044
1045 hp = tcp_get_md5sig_pool();
1046 if (!hp)
1047 goto clear_hash_noput;
1048 desc = &hp->md5_desc;
1049
1050 if (crypto_hash_init(desc))
1051 goto clear_hash;
1052 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1053 goto clear_hash;
1054 if (tcp_md5_hash_header(hp, th))
1055 goto clear_hash;
1056 if (tcp_md5_hash_key(hp, key))
1057 goto clear_hash;
1058 if (crypto_hash_final(desc, md5_hash))
1059 goto clear_hash;
1060
1061 tcp_put_md5sig_pool();
1062 return 0;
1063
1064 clear_hash:
1065 tcp_put_md5sig_pool();
1066 clear_hash_noput:
1067 memset(md5_hash, 0, 16);
1068 return 1;
1069 }
1070
1071 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1072 const struct sock *sk,
1073 const struct sk_buff *skb)
1074 {
1075 struct tcp_md5sig_pool *hp;
1076 struct hash_desc *desc;
1077 const struct tcphdr *th = tcp_hdr(skb);
1078 __be32 saddr, daddr;
1079
1080 if (sk) { /* valid for establish/request sockets */
1081 saddr = sk->sk_rcv_saddr;
1082 daddr = sk->sk_daddr;
1083 } else {
1084 const struct iphdr *iph = ip_hdr(skb);
1085 saddr = iph->saddr;
1086 daddr = iph->daddr;
1087 }
1088
1089 hp = tcp_get_md5sig_pool();
1090 if (!hp)
1091 goto clear_hash_noput;
1092 desc = &hp->md5_desc;
1093
1094 if (crypto_hash_init(desc))
1095 goto clear_hash;
1096
1097 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1098 goto clear_hash;
1099 if (tcp_md5_hash_header(hp, th))
1100 goto clear_hash;
1101 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1102 goto clear_hash;
1103 if (tcp_md5_hash_key(hp, key))
1104 goto clear_hash;
1105 if (crypto_hash_final(desc, md5_hash))
1106 goto clear_hash;
1107
1108 tcp_put_md5sig_pool();
1109 return 0;
1110
1111 clear_hash:
1112 tcp_put_md5sig_pool();
1113 clear_hash_noput:
1114 memset(md5_hash, 0, 16);
1115 return 1;
1116 }
1117 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1118
1119 #endif
1120
1121 /* Called with rcu_read_lock() */
1122 static bool tcp_v4_inbound_md5_hash(const struct sock *sk,
1123 const struct sk_buff *skb)
1124 {
1125 #ifdef CONFIG_TCP_MD5SIG
1126 /*
1127 * This gets called for each TCP segment that arrives
1128 * so we want to be efficient.
1129 * We have 3 drop cases:
1130 * o No MD5 hash and one expected.
1131 * o MD5 hash and we're not expecting one.
1132 * o MD5 hash and its wrong.
1133 */
1134 const __u8 *hash_location = NULL;
1135 struct tcp_md5sig_key *hash_expected;
1136 const struct iphdr *iph = ip_hdr(skb);
1137 const struct tcphdr *th = tcp_hdr(skb);
1138 int genhash;
1139 unsigned char newhash[16];
1140
1141 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1142 AF_INET);
1143 hash_location = tcp_parse_md5sig_option(th);
1144
1145 /* We've parsed the options - do we have a hash? */
1146 if (!hash_expected && !hash_location)
1147 return false;
1148
1149 if (hash_expected && !hash_location) {
1150 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1151 return true;
1152 }
1153
1154 if (!hash_expected && hash_location) {
1155 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1156 return true;
1157 }
1158
1159 /* Okay, so this is hash_expected and hash_location -
1160 * so we need to calculate the checksum.
1161 */
1162 genhash = tcp_v4_md5_hash_skb(newhash,
1163 hash_expected,
1164 NULL, skb);
1165
1166 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1167 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1168 &iph->saddr, ntohs(th->source),
1169 &iph->daddr, ntohs(th->dest),
1170 genhash ? " tcp_v4_calc_md5_hash failed"
1171 : "");
1172 return true;
1173 }
1174 return false;
1175 #endif
1176 return false;
1177 }
1178
1179 static void tcp_v4_init_req(struct request_sock *req,
1180 const struct sock *sk_listener,
1181 struct sk_buff *skb)
1182 {
1183 struct inet_request_sock *ireq = inet_rsk(req);
1184
1185 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1186 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1187 ireq->no_srccheck = inet_sk(sk_listener)->transparent;
1188 ireq->opt = tcp_v4_save_options(skb);
1189 }
1190
1191 static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
1192 struct flowi *fl,
1193 const struct request_sock *req,
1194 bool *strict)
1195 {
1196 struct dst_entry *dst = inet_csk_route_req(sk, &fl->u.ip4, req);
1197
1198 if (strict) {
1199 if (fl->u.ip4.daddr == inet_rsk(req)->ir_rmt_addr)
1200 *strict = true;
1201 else
1202 *strict = false;
1203 }
1204
1205 return dst;
1206 }
1207
1208 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1209 .family = PF_INET,
1210 .obj_size = sizeof(struct tcp_request_sock),
1211 .rtx_syn_ack = tcp_rtx_synack,
1212 .send_ack = tcp_v4_reqsk_send_ack,
1213 .destructor = tcp_v4_reqsk_destructor,
1214 .send_reset = tcp_v4_send_reset,
1215 .syn_ack_timeout = tcp_syn_ack_timeout,
1216 };
1217
1218 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1219 .mss_clamp = TCP_MSS_DEFAULT,
1220 #ifdef CONFIG_TCP_MD5SIG
1221 .req_md5_lookup = tcp_v4_md5_lookup,
1222 .calc_md5_hash = tcp_v4_md5_hash_skb,
1223 #endif
1224 .init_req = tcp_v4_init_req,
1225 #ifdef CONFIG_SYN_COOKIES
1226 .cookie_init_seq = cookie_v4_init_sequence,
1227 #endif
1228 .route_req = tcp_v4_route_req,
1229 .init_seq = tcp_v4_init_sequence,
1230 .send_synack = tcp_v4_send_synack,
1231 };
1232
1233 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1234 {
1235 /* Never answer to SYNs send to broadcast or multicast */
1236 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1237 goto drop;
1238
1239 return tcp_conn_request(&tcp_request_sock_ops,
1240 &tcp_request_sock_ipv4_ops, sk, skb);
1241
1242 drop:
1243 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1244 return 0;
1245 }
1246 EXPORT_SYMBOL(tcp_v4_conn_request);
1247
1248
1249 /*
1250 * The three way handshake has completed - we got a valid synack -
1251 * now create the new socket.
1252 */
1253 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1254 struct request_sock *req,
1255 struct dst_entry *dst,
1256 struct request_sock *req_unhash,
1257 bool *own_req)
1258 {
1259 struct inet_request_sock *ireq;
1260 struct inet_sock *newinet;
1261 struct tcp_sock *newtp;
1262 struct sock *newsk;
1263 #ifdef CONFIG_TCP_MD5SIG
1264 struct tcp_md5sig_key *key;
1265 #endif
1266 struct ip_options_rcu *inet_opt;
1267
1268 if (sk_acceptq_is_full(sk))
1269 goto exit_overflow;
1270
1271 newsk = tcp_create_openreq_child(sk, req, skb);
1272 if (!newsk)
1273 goto exit_nonewsk;
1274
1275 newsk->sk_gso_type = SKB_GSO_TCPV4;
1276 inet_sk_rx_dst_set(newsk, skb);
1277
1278 newtp = tcp_sk(newsk);
1279 newinet = inet_sk(newsk);
1280 ireq = inet_rsk(req);
1281 sk_daddr_set(newsk, ireq->ir_rmt_addr);
1282 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1283 newsk->sk_bound_dev_if = ireq->ir_iif;
1284 newinet->inet_saddr = ireq->ir_loc_addr;
1285 inet_opt = ireq->opt;
1286 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1287 ireq->opt = NULL;
1288 newinet->mc_index = inet_iif(skb);
1289 newinet->mc_ttl = ip_hdr(skb)->ttl;
1290 newinet->rcv_tos = ip_hdr(skb)->tos;
1291 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1292 if (inet_opt)
1293 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1294 newinet->inet_id = newtp->write_seq ^ jiffies;
1295
1296 if (!dst) {
1297 dst = inet_csk_route_child_sock(sk, newsk, req);
1298 if (!dst)
1299 goto put_and_exit;
1300 } else {
1301 /* syncookie case : see end of cookie_v4_check() */
1302 }
1303 sk_setup_caps(newsk, dst);
1304
1305 tcp_ca_openreq_child(newsk, dst);
1306
1307 tcp_sync_mss(newsk, dst_mtu(dst));
1308 newtp->advmss = dst_metric_advmss(dst);
1309 if (tcp_sk(sk)->rx_opt.user_mss &&
1310 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1311 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1312
1313 tcp_initialize_rcv_mss(newsk);
1314
1315 #ifdef CONFIG_TCP_MD5SIG
1316 /* Copy over the MD5 key from the original socket */
1317 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1318 AF_INET);
1319 if (key) {
1320 /*
1321 * We're using one, so create a matching key
1322 * on the newsk structure. If we fail to get
1323 * memory, then we end up not copying the key
1324 * across. Shucks.
1325 */
1326 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1327 AF_INET, key->key, key->keylen, GFP_ATOMIC);
1328 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1329 }
1330 #endif
1331
1332 if (__inet_inherit_port(sk, newsk) < 0)
1333 goto put_and_exit;
1334 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1335 if (*own_req)
1336 tcp_move_syn(newtp, req);
1337
1338 return newsk;
1339
1340 exit_overflow:
1341 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1342 exit_nonewsk:
1343 dst_release(dst);
1344 exit:
1345 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1346 return NULL;
1347 put_and_exit:
1348 inet_csk_prepare_forced_close(newsk);
1349 tcp_done(newsk);
1350 goto exit;
1351 }
1352 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1353
1354 static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb)
1355 {
1356 #ifdef CONFIG_SYN_COOKIES
1357 const struct tcphdr *th = tcp_hdr(skb);
1358
1359 if (!th->syn)
1360 sk = cookie_v4_check(sk, skb);
1361 #endif
1362 return sk;
1363 }
1364
1365 /* The socket must have it's spinlock held when we get
1366 * here, unless it is a TCP_LISTEN socket.
1367 *
1368 * We have a potential double-lock case here, so even when
1369 * doing backlog processing we use the BH locking scheme.
1370 * This is because we cannot sleep with the original spinlock
1371 * held.
1372 */
1373 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1374 {
1375 struct sock *rsk;
1376
1377 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1378 struct dst_entry *dst = sk->sk_rx_dst;
1379
1380 sock_rps_save_rxhash(sk, skb);
1381 sk_mark_napi_id(sk, skb);
1382 if (dst) {
1383 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1384 !dst->ops->check(dst, 0)) {
1385 dst_release(dst);
1386 sk->sk_rx_dst = NULL;
1387 }
1388 }
1389 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1390 return 0;
1391 }
1392
1393 if (tcp_checksum_complete(skb))
1394 goto csum_err;
1395
1396 if (sk->sk_state == TCP_LISTEN) {
1397 struct sock *nsk = tcp_v4_cookie_check(sk, skb);
1398
1399 if (!nsk)
1400 goto discard;
1401 if (nsk != sk) {
1402 sock_rps_save_rxhash(nsk, skb);
1403 sk_mark_napi_id(nsk, skb);
1404 if (tcp_child_process(sk, nsk, skb)) {
1405 rsk = nsk;
1406 goto reset;
1407 }
1408 return 0;
1409 }
1410 } else
1411 sock_rps_save_rxhash(sk, skb);
1412
1413 if (tcp_rcv_state_process(sk, skb)) {
1414 rsk = sk;
1415 goto reset;
1416 }
1417 return 0;
1418
1419 reset:
1420 tcp_v4_send_reset(rsk, skb);
1421 discard:
1422 kfree_skb(skb);
1423 /* Be careful here. If this function gets more complicated and
1424 * gcc suffers from register pressure on the x86, sk (in %ebx)
1425 * might be destroyed here. This current version compiles correctly,
1426 * but you have been warned.
1427 */
1428 return 0;
1429
1430 csum_err:
1431 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1432 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1433 goto discard;
1434 }
1435 EXPORT_SYMBOL(tcp_v4_do_rcv);
1436
1437 void tcp_v4_early_demux(struct sk_buff *skb)
1438 {
1439 const struct iphdr *iph;
1440 const struct tcphdr *th;
1441 struct sock *sk;
1442
1443 if (skb->pkt_type != PACKET_HOST)
1444 return;
1445
1446 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1447 return;
1448
1449 iph = ip_hdr(skb);
1450 th = tcp_hdr(skb);
1451
1452 if (th->doff < sizeof(struct tcphdr) / 4)
1453 return;
1454
1455 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1456 iph->saddr, th->source,
1457 iph->daddr, ntohs(th->dest),
1458 skb->skb_iif);
1459 if (sk) {
1460 skb->sk = sk;
1461 skb->destructor = sock_edemux;
1462 if (sk_fullsock(sk)) {
1463 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1464
1465 if (dst)
1466 dst = dst_check(dst, 0);
1467 if (dst &&
1468 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1469 skb_dst_set_noref(skb, dst);
1470 }
1471 }
1472 }
1473
1474 /* Packet is added to VJ-style prequeue for processing in process
1475 * context, if a reader task is waiting. Apparently, this exciting
1476 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1477 * failed somewhere. Latency? Burstiness? Well, at least now we will
1478 * see, why it failed. 8)8) --ANK
1479 *
1480 */
1481 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1482 {
1483 struct tcp_sock *tp = tcp_sk(sk);
1484
1485 if (sysctl_tcp_low_latency || !tp->ucopy.task)
1486 return false;
1487
1488 if (skb->len <= tcp_hdrlen(skb) &&
1489 skb_queue_len(&tp->ucopy.prequeue) == 0)
1490 return false;
1491
1492 /* Before escaping RCU protected region, we need to take care of skb
1493 * dst. Prequeue is only enabled for established sockets.
1494 * For such sockets, we might need the skb dst only to set sk->sk_rx_dst
1495 * Instead of doing full sk_rx_dst validity here, let's perform
1496 * an optimistic check.
1497 */
1498 if (likely(sk->sk_rx_dst))
1499 skb_dst_drop(skb);
1500 else
1501 skb_dst_force_safe(skb);
1502
1503 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1504 tp->ucopy.memory += skb->truesize;
1505 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1506 struct sk_buff *skb1;
1507
1508 BUG_ON(sock_owned_by_user(sk));
1509
1510 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1511 sk_backlog_rcv(sk, skb1);
1512 NET_INC_STATS_BH(sock_net(sk),
1513 LINUX_MIB_TCPPREQUEUEDROPPED);
1514 }
1515
1516 tp->ucopy.memory = 0;
1517 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1518 wake_up_interruptible_sync_poll(sk_sleep(sk),
1519 POLLIN | POLLRDNORM | POLLRDBAND);
1520 if (!inet_csk_ack_scheduled(sk))
1521 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1522 (3 * tcp_rto_min(sk)) / 4,
1523 TCP_RTO_MAX);
1524 }
1525 return true;
1526 }
1527 EXPORT_SYMBOL(tcp_prequeue);
1528
1529 /*
1530 * From tcp_input.c
1531 */
1532
1533 int tcp_v4_rcv(struct sk_buff *skb)
1534 {
1535 const struct iphdr *iph;
1536 const struct tcphdr *th;
1537 struct sock *sk;
1538 int ret;
1539 struct net *net = dev_net(skb->dev);
1540
1541 if (skb->pkt_type != PACKET_HOST)
1542 goto discard_it;
1543
1544 /* Count it even if it's bad */
1545 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1546
1547 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1548 goto discard_it;
1549
1550 th = tcp_hdr(skb);
1551
1552 if (th->doff < sizeof(struct tcphdr) / 4)
1553 goto bad_packet;
1554 if (!pskb_may_pull(skb, th->doff * 4))
1555 goto discard_it;
1556
1557 /* An explanation is required here, I think.
1558 * Packet length and doff are validated by header prediction,
1559 * provided case of th->doff==0 is eliminated.
1560 * So, we defer the checks. */
1561
1562 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
1563 goto csum_error;
1564
1565 th = tcp_hdr(skb);
1566 iph = ip_hdr(skb);
1567 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1568 * barrier() makes sure compiler wont play fool^Waliasing games.
1569 */
1570 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1571 sizeof(struct inet_skb_parm));
1572 barrier();
1573
1574 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1575 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1576 skb->len - th->doff * 4);
1577 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1578 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1579 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1580 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1581 TCP_SKB_CB(skb)->sacked = 0;
1582
1583 lookup:
1584 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1585 if (!sk)
1586 goto no_tcp_socket;
1587
1588 process:
1589 if (sk->sk_state == TCP_TIME_WAIT)
1590 goto do_time_wait;
1591
1592 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1593 struct request_sock *req = inet_reqsk(sk);
1594 struct sock *nsk = NULL;
1595
1596 sk = req->rsk_listener;
1597 if (tcp_v4_inbound_md5_hash(sk, skb))
1598 goto discard_and_relse;
1599 if (likely(sk->sk_state == TCP_LISTEN)) {
1600 nsk = tcp_check_req(sk, skb, req, false);
1601 } else {
1602 inet_csk_reqsk_queue_drop_and_put(sk, req);
1603 goto lookup;
1604 }
1605 if (!nsk) {
1606 reqsk_put(req);
1607 goto discard_it;
1608 }
1609 if (nsk == sk) {
1610 sock_hold(sk);
1611 reqsk_put(req);
1612 } else if (tcp_child_process(sk, nsk, skb)) {
1613 tcp_v4_send_reset(nsk, skb);
1614 goto discard_it;
1615 } else {
1616 return 0;
1617 }
1618 }
1619 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1620 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1621 goto discard_and_relse;
1622 }
1623
1624 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1625 goto discard_and_relse;
1626
1627 if (tcp_v4_inbound_md5_hash(sk, skb))
1628 goto discard_and_relse;
1629
1630 nf_reset(skb);
1631
1632 if (sk_filter(sk, skb))
1633 goto discard_and_relse;
1634
1635 skb->dev = NULL;
1636
1637 if (sk->sk_state == TCP_LISTEN) {
1638 ret = tcp_v4_do_rcv(sk, skb);
1639 goto put_and_return;
1640 }
1641
1642 sk_incoming_cpu_update(sk);
1643
1644 bh_lock_sock_nested(sk);
1645 tcp_sk(sk)->segs_in += max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1646 ret = 0;
1647 if (!sock_owned_by_user(sk)) {
1648 if (!tcp_prequeue(sk, skb))
1649 ret = tcp_v4_do_rcv(sk, skb);
1650 } else if (unlikely(sk_add_backlog(sk, skb,
1651 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1652 bh_unlock_sock(sk);
1653 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1654 goto discard_and_relse;
1655 }
1656 bh_unlock_sock(sk);
1657
1658 put_and_return:
1659 sock_put(sk);
1660
1661 return ret;
1662
1663 no_tcp_socket:
1664 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1665 goto discard_it;
1666
1667 if (tcp_checksum_complete(skb)) {
1668 csum_error:
1669 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1670 bad_packet:
1671 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1672 } else {
1673 tcp_v4_send_reset(NULL, skb);
1674 }
1675
1676 discard_it:
1677 /* Discard frame. */
1678 kfree_skb(skb);
1679 return 0;
1680
1681 discard_and_relse:
1682 sock_put(sk);
1683 goto discard_it;
1684
1685 do_time_wait:
1686 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1687 inet_twsk_put(inet_twsk(sk));
1688 goto discard_it;
1689 }
1690
1691 if (tcp_checksum_complete(skb)) {
1692 inet_twsk_put(inet_twsk(sk));
1693 goto csum_error;
1694 }
1695 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1696 case TCP_TW_SYN: {
1697 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1698 &tcp_hashinfo,
1699 iph->saddr, th->source,
1700 iph->daddr, th->dest,
1701 inet_iif(skb));
1702 if (sk2) {
1703 inet_twsk_deschedule_put(inet_twsk(sk));
1704 sk = sk2;
1705 goto process;
1706 }
1707 /* Fall through to ACK */
1708 }
1709 case TCP_TW_ACK:
1710 tcp_v4_timewait_ack(sk, skb);
1711 break;
1712 case TCP_TW_RST:
1713 tcp_v4_send_reset(sk, skb);
1714 inet_twsk_deschedule_put(inet_twsk(sk));
1715 goto discard_it;
1716 case TCP_TW_SUCCESS:;
1717 }
1718 goto discard_it;
1719 }
1720
1721 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1722 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1723 .twsk_unique = tcp_twsk_unique,
1724 .twsk_destructor= tcp_twsk_destructor,
1725 };
1726
1727 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
1728 {
1729 struct dst_entry *dst = skb_dst(skb);
1730
1731 if (dst && dst_hold_safe(dst)) {
1732 sk->sk_rx_dst = dst;
1733 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
1734 }
1735 }
1736 EXPORT_SYMBOL(inet_sk_rx_dst_set);
1737
1738 const struct inet_connection_sock_af_ops ipv4_specific = {
1739 .queue_xmit = ip_queue_xmit,
1740 .send_check = tcp_v4_send_check,
1741 .rebuild_header = inet_sk_rebuild_header,
1742 .sk_rx_dst_set = inet_sk_rx_dst_set,
1743 .conn_request = tcp_v4_conn_request,
1744 .syn_recv_sock = tcp_v4_syn_recv_sock,
1745 .net_header_len = sizeof(struct iphdr),
1746 .setsockopt = ip_setsockopt,
1747 .getsockopt = ip_getsockopt,
1748 .addr2sockaddr = inet_csk_addr2sockaddr,
1749 .sockaddr_len = sizeof(struct sockaddr_in),
1750 .bind_conflict = inet_csk_bind_conflict,
1751 #ifdef CONFIG_COMPAT
1752 .compat_setsockopt = compat_ip_setsockopt,
1753 .compat_getsockopt = compat_ip_getsockopt,
1754 #endif
1755 .mtu_reduced = tcp_v4_mtu_reduced,
1756 };
1757 EXPORT_SYMBOL(ipv4_specific);
1758
1759 #ifdef CONFIG_TCP_MD5SIG
1760 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1761 .md5_lookup = tcp_v4_md5_lookup,
1762 .calc_md5_hash = tcp_v4_md5_hash_skb,
1763 .md5_parse = tcp_v4_parse_md5_keys,
1764 };
1765 #endif
1766
1767 /* NOTE: A lot of things set to zero explicitly by call to
1768 * sk_alloc() so need not be done here.
1769 */
1770 static int tcp_v4_init_sock(struct sock *sk)
1771 {
1772 struct inet_connection_sock *icsk = inet_csk(sk);
1773
1774 tcp_init_sock(sk);
1775
1776 icsk->icsk_af_ops = &ipv4_specific;
1777
1778 #ifdef CONFIG_TCP_MD5SIG
1779 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
1780 #endif
1781
1782 return 0;
1783 }
1784
1785 void tcp_v4_destroy_sock(struct sock *sk)
1786 {
1787 struct tcp_sock *tp = tcp_sk(sk);
1788
1789 tcp_clear_xmit_timers(sk);
1790
1791 tcp_cleanup_congestion_control(sk);
1792
1793 /* Cleanup up the write buffer. */
1794 tcp_write_queue_purge(sk);
1795
1796 /* Cleans up our, hopefully empty, out_of_order_queue. */
1797 __skb_queue_purge(&tp->out_of_order_queue);
1798
1799 #ifdef CONFIG_TCP_MD5SIG
1800 /* Clean up the MD5 key list, if any */
1801 if (tp->md5sig_info) {
1802 tcp_clear_md5_list(sk);
1803 kfree_rcu(tp->md5sig_info, rcu);
1804 tp->md5sig_info = NULL;
1805 }
1806 #endif
1807
1808 /* Clean prequeue, it must be empty really */
1809 __skb_queue_purge(&tp->ucopy.prequeue);
1810
1811 /* Clean up a referenced TCP bind bucket. */
1812 if (inet_csk(sk)->icsk_bind_hash)
1813 inet_put_port(sk);
1814
1815 BUG_ON(tp->fastopen_rsk);
1816
1817 /* If socket is aborted during connect operation */
1818 tcp_free_fastopen_req(tp);
1819 tcp_saved_syn_free(tp);
1820
1821 sk_sockets_allocated_dec(sk);
1822
1823 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
1824 sock_release_memcg(sk);
1825 }
1826 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1827
1828 #ifdef CONFIG_PROC_FS
1829 /* Proc filesystem TCP sock list dumping. */
1830
1831 /*
1832 * Get next listener socket follow cur. If cur is NULL, get first socket
1833 * starting from bucket given in st->bucket; when st->bucket is zero the
1834 * very first socket in the hash table is returned.
1835 */
1836 static void *listening_get_next(struct seq_file *seq, void *cur)
1837 {
1838 struct inet_connection_sock *icsk;
1839 struct hlist_nulls_node *node;
1840 struct sock *sk = cur;
1841 struct inet_listen_hashbucket *ilb;
1842 struct tcp_iter_state *st = seq->private;
1843 struct net *net = seq_file_net(seq);
1844
1845 if (!sk) {
1846 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1847 spin_lock_bh(&ilb->lock);
1848 sk = sk_nulls_head(&ilb->head);
1849 st->offset = 0;
1850 goto get_sk;
1851 }
1852 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1853 ++st->num;
1854 ++st->offset;
1855
1856 sk = sk_nulls_next(sk);
1857 get_sk:
1858 sk_nulls_for_each_from(sk, node) {
1859 if (!net_eq(sock_net(sk), net))
1860 continue;
1861 if (sk->sk_family == st->family) {
1862 cur = sk;
1863 goto out;
1864 }
1865 icsk = inet_csk(sk);
1866 }
1867 spin_unlock_bh(&ilb->lock);
1868 st->offset = 0;
1869 if (++st->bucket < INET_LHTABLE_SIZE) {
1870 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1871 spin_lock_bh(&ilb->lock);
1872 sk = sk_nulls_head(&ilb->head);
1873 goto get_sk;
1874 }
1875 cur = NULL;
1876 out:
1877 return cur;
1878 }
1879
1880 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1881 {
1882 struct tcp_iter_state *st = seq->private;
1883 void *rc;
1884
1885 st->bucket = 0;
1886 st->offset = 0;
1887 rc = listening_get_next(seq, NULL);
1888
1889 while (rc && *pos) {
1890 rc = listening_get_next(seq, rc);
1891 --*pos;
1892 }
1893 return rc;
1894 }
1895
1896 static inline bool empty_bucket(const struct tcp_iter_state *st)
1897 {
1898 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
1899 }
1900
1901 /*
1902 * Get first established socket starting from bucket given in st->bucket.
1903 * If st->bucket is zero, the very first socket in the hash is returned.
1904 */
1905 static void *established_get_first(struct seq_file *seq)
1906 {
1907 struct tcp_iter_state *st = seq->private;
1908 struct net *net = seq_file_net(seq);
1909 void *rc = NULL;
1910
1911 st->offset = 0;
1912 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
1913 struct sock *sk;
1914 struct hlist_nulls_node *node;
1915 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
1916
1917 /* Lockless fast path for the common case of empty buckets */
1918 if (empty_bucket(st))
1919 continue;
1920
1921 spin_lock_bh(lock);
1922 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
1923 if (sk->sk_family != st->family ||
1924 !net_eq(sock_net(sk), net)) {
1925 continue;
1926 }
1927 rc = sk;
1928 goto out;
1929 }
1930 spin_unlock_bh(lock);
1931 }
1932 out:
1933 return rc;
1934 }
1935
1936 static void *established_get_next(struct seq_file *seq, void *cur)
1937 {
1938 struct sock *sk = cur;
1939 struct hlist_nulls_node *node;
1940 struct tcp_iter_state *st = seq->private;
1941 struct net *net = seq_file_net(seq);
1942
1943 ++st->num;
1944 ++st->offset;
1945
1946 sk = sk_nulls_next(sk);
1947
1948 sk_nulls_for_each_from(sk, node) {
1949 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
1950 return sk;
1951 }
1952
1953 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
1954 ++st->bucket;
1955 return established_get_first(seq);
1956 }
1957
1958 static void *established_get_idx(struct seq_file *seq, loff_t pos)
1959 {
1960 struct tcp_iter_state *st = seq->private;
1961 void *rc;
1962
1963 st->bucket = 0;
1964 rc = established_get_first(seq);
1965
1966 while (rc && pos) {
1967 rc = established_get_next(seq, rc);
1968 --pos;
1969 }
1970 return rc;
1971 }
1972
1973 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
1974 {
1975 void *rc;
1976 struct tcp_iter_state *st = seq->private;
1977
1978 st->state = TCP_SEQ_STATE_LISTENING;
1979 rc = listening_get_idx(seq, &pos);
1980
1981 if (!rc) {
1982 st->state = TCP_SEQ_STATE_ESTABLISHED;
1983 rc = established_get_idx(seq, pos);
1984 }
1985
1986 return rc;
1987 }
1988
1989 static void *tcp_seek_last_pos(struct seq_file *seq)
1990 {
1991 struct tcp_iter_state *st = seq->private;
1992 int offset = st->offset;
1993 int orig_num = st->num;
1994 void *rc = NULL;
1995
1996 switch (st->state) {
1997 case TCP_SEQ_STATE_LISTENING:
1998 if (st->bucket >= INET_LHTABLE_SIZE)
1999 break;
2000 st->state = TCP_SEQ_STATE_LISTENING;
2001 rc = listening_get_next(seq, NULL);
2002 while (offset-- && rc)
2003 rc = listening_get_next(seq, rc);
2004 if (rc)
2005 break;
2006 st->bucket = 0;
2007 st->state = TCP_SEQ_STATE_ESTABLISHED;
2008 /* Fallthrough */
2009 case TCP_SEQ_STATE_ESTABLISHED:
2010 if (st->bucket > tcp_hashinfo.ehash_mask)
2011 break;
2012 rc = established_get_first(seq);
2013 while (offset-- && rc)
2014 rc = established_get_next(seq, rc);
2015 }
2016
2017 st->num = orig_num;
2018
2019 return rc;
2020 }
2021
2022 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2023 {
2024 struct tcp_iter_state *st = seq->private;
2025 void *rc;
2026
2027 if (*pos && *pos == st->last_pos) {
2028 rc = tcp_seek_last_pos(seq);
2029 if (rc)
2030 goto out;
2031 }
2032
2033 st->state = TCP_SEQ_STATE_LISTENING;
2034 st->num = 0;
2035 st->bucket = 0;
2036 st->offset = 0;
2037 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2038
2039 out:
2040 st->last_pos = *pos;
2041 return rc;
2042 }
2043
2044 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2045 {
2046 struct tcp_iter_state *st = seq->private;
2047 void *rc = NULL;
2048
2049 if (v == SEQ_START_TOKEN) {
2050 rc = tcp_get_idx(seq, 0);
2051 goto out;
2052 }
2053
2054 switch (st->state) {
2055 case TCP_SEQ_STATE_LISTENING:
2056 rc = listening_get_next(seq, v);
2057 if (!rc) {
2058 st->state = TCP_SEQ_STATE_ESTABLISHED;
2059 st->bucket = 0;
2060 st->offset = 0;
2061 rc = established_get_first(seq);
2062 }
2063 break;
2064 case TCP_SEQ_STATE_ESTABLISHED:
2065 rc = established_get_next(seq, v);
2066 break;
2067 }
2068 out:
2069 ++*pos;
2070 st->last_pos = *pos;
2071 return rc;
2072 }
2073
2074 static void tcp_seq_stop(struct seq_file *seq, void *v)
2075 {
2076 struct tcp_iter_state *st = seq->private;
2077
2078 switch (st->state) {
2079 case TCP_SEQ_STATE_LISTENING:
2080 if (v != SEQ_START_TOKEN)
2081 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2082 break;
2083 case TCP_SEQ_STATE_ESTABLISHED:
2084 if (v)
2085 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2086 break;
2087 }
2088 }
2089
2090 int tcp_seq_open(struct inode *inode, struct file *file)
2091 {
2092 struct tcp_seq_afinfo *afinfo = PDE_DATA(inode);
2093 struct tcp_iter_state *s;
2094 int err;
2095
2096 err = seq_open_net(inode, file, &afinfo->seq_ops,
2097 sizeof(struct tcp_iter_state));
2098 if (err < 0)
2099 return err;
2100
2101 s = ((struct seq_file *)file->private_data)->private;
2102 s->family = afinfo->family;
2103 s->last_pos = 0;
2104 return 0;
2105 }
2106 EXPORT_SYMBOL(tcp_seq_open);
2107
2108 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2109 {
2110 int rc = 0;
2111 struct proc_dir_entry *p;
2112
2113 afinfo->seq_ops.start = tcp_seq_start;
2114 afinfo->seq_ops.next = tcp_seq_next;
2115 afinfo->seq_ops.stop = tcp_seq_stop;
2116
2117 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2118 afinfo->seq_fops, afinfo);
2119 if (!p)
2120 rc = -ENOMEM;
2121 return rc;
2122 }
2123 EXPORT_SYMBOL(tcp_proc_register);
2124
2125 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2126 {
2127 remove_proc_entry(afinfo->name, net->proc_net);
2128 }
2129 EXPORT_SYMBOL(tcp_proc_unregister);
2130
2131 static void get_openreq4(const struct request_sock *req,
2132 struct seq_file *f, int i)
2133 {
2134 const struct inet_request_sock *ireq = inet_rsk(req);
2135 long delta = req->rsk_timer.expires - jiffies;
2136
2137 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2138 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2139 i,
2140 ireq->ir_loc_addr,
2141 ireq->ir_num,
2142 ireq->ir_rmt_addr,
2143 ntohs(ireq->ir_rmt_port),
2144 TCP_SYN_RECV,
2145 0, 0, /* could print option size, but that is af dependent. */
2146 1, /* timers active (only the expire timer) */
2147 jiffies_delta_to_clock_t(delta),
2148 req->num_timeout,
2149 from_kuid_munged(seq_user_ns(f),
2150 sock_i_uid(req->rsk_listener)),
2151 0, /* non standard timer */
2152 0, /* open_requests have no inode */
2153 0,
2154 req);
2155 }
2156
2157 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2158 {
2159 int timer_active;
2160 unsigned long timer_expires;
2161 const struct tcp_sock *tp = tcp_sk(sk);
2162 const struct inet_connection_sock *icsk = inet_csk(sk);
2163 const struct inet_sock *inet = inet_sk(sk);
2164 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2165 __be32 dest = inet->inet_daddr;
2166 __be32 src = inet->inet_rcv_saddr;
2167 __u16 destp = ntohs(inet->inet_dport);
2168 __u16 srcp = ntohs(inet->inet_sport);
2169 int rx_queue;
2170 int state;
2171
2172 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2173 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
2174 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2175 timer_active = 1;
2176 timer_expires = icsk->icsk_timeout;
2177 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2178 timer_active = 4;
2179 timer_expires = icsk->icsk_timeout;
2180 } else if (timer_pending(&sk->sk_timer)) {
2181 timer_active = 2;
2182 timer_expires = sk->sk_timer.expires;
2183 } else {
2184 timer_active = 0;
2185 timer_expires = jiffies;
2186 }
2187
2188 state = sk_state_load(sk);
2189 if (state == TCP_LISTEN)
2190 rx_queue = sk->sk_ack_backlog;
2191 else
2192 /* Because we don't lock the socket,
2193 * we might find a transient negative value.
2194 */
2195 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2196
2197 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2198 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2199 i, src, srcp, dest, destp, state,
2200 tp->write_seq - tp->snd_una,
2201 rx_queue,
2202 timer_active,
2203 jiffies_delta_to_clock_t(timer_expires - jiffies),
2204 icsk->icsk_retransmits,
2205 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2206 icsk->icsk_probes_out,
2207 sock_i_ino(sk),
2208 atomic_read(&sk->sk_refcnt), sk,
2209 jiffies_to_clock_t(icsk->icsk_rto),
2210 jiffies_to_clock_t(icsk->icsk_ack.ato),
2211 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2212 tp->snd_cwnd,
2213 state == TCP_LISTEN ?
2214 fastopenq->max_qlen :
2215 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2216 }
2217
2218 static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2219 struct seq_file *f, int i)
2220 {
2221 long delta = tw->tw_timer.expires - jiffies;
2222 __be32 dest, src;
2223 __u16 destp, srcp;
2224
2225 dest = tw->tw_daddr;
2226 src = tw->tw_rcv_saddr;
2227 destp = ntohs(tw->tw_dport);
2228 srcp = ntohs(tw->tw_sport);
2229
2230 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2231 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2232 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2233 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2234 atomic_read(&tw->tw_refcnt), tw);
2235 }
2236
2237 #define TMPSZ 150
2238
2239 static int tcp4_seq_show(struct seq_file *seq, void *v)
2240 {
2241 struct tcp_iter_state *st;
2242 struct sock *sk = v;
2243
2244 seq_setwidth(seq, TMPSZ - 1);
2245 if (v == SEQ_START_TOKEN) {
2246 seq_puts(seq, " sl local_address rem_address st tx_queue "
2247 "rx_queue tr tm->when retrnsmt uid timeout "
2248 "inode");
2249 goto out;
2250 }
2251 st = seq->private;
2252
2253 if (sk->sk_state == TCP_TIME_WAIT)
2254 get_timewait4_sock(v, seq, st->num);
2255 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2256 get_openreq4(v, seq, st->num);
2257 else
2258 get_tcp4_sock(v, seq, st->num);
2259 out:
2260 seq_pad(seq, '\n');
2261 return 0;
2262 }
2263
2264 static const struct file_operations tcp_afinfo_seq_fops = {
2265 .owner = THIS_MODULE,
2266 .open = tcp_seq_open,
2267 .read = seq_read,
2268 .llseek = seq_lseek,
2269 .release = seq_release_net
2270 };
2271
2272 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2273 .name = "tcp",
2274 .family = AF_INET,
2275 .seq_fops = &tcp_afinfo_seq_fops,
2276 .seq_ops = {
2277 .show = tcp4_seq_show,
2278 },
2279 };
2280
2281 static int __net_init tcp4_proc_init_net(struct net *net)
2282 {
2283 return tcp_proc_register(net, &tcp4_seq_afinfo);
2284 }
2285
2286 static void __net_exit tcp4_proc_exit_net(struct net *net)
2287 {
2288 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2289 }
2290
2291 static struct pernet_operations tcp4_net_ops = {
2292 .init = tcp4_proc_init_net,
2293 .exit = tcp4_proc_exit_net,
2294 };
2295
2296 int __init tcp4_proc_init(void)
2297 {
2298 return register_pernet_subsys(&tcp4_net_ops);
2299 }
2300
2301 void tcp4_proc_exit(void)
2302 {
2303 unregister_pernet_subsys(&tcp4_net_ops);
2304 }
2305 #endif /* CONFIG_PROC_FS */
2306
2307 struct proto tcp_prot = {
2308 .name = "TCP",
2309 .owner = THIS_MODULE,
2310 .close = tcp_close,
2311 .connect = tcp_v4_connect,
2312 .disconnect = tcp_disconnect,
2313 .accept = inet_csk_accept,
2314 .ioctl = tcp_ioctl,
2315 .init = tcp_v4_init_sock,
2316 .destroy = tcp_v4_destroy_sock,
2317 .shutdown = tcp_shutdown,
2318 .setsockopt = tcp_setsockopt,
2319 .getsockopt = tcp_getsockopt,
2320 .recvmsg = tcp_recvmsg,
2321 .sendmsg = tcp_sendmsg,
2322 .sendpage = tcp_sendpage,
2323 .backlog_rcv = tcp_v4_do_rcv,
2324 .release_cb = tcp_release_cb,
2325 .hash = inet_hash,
2326 .unhash = inet_unhash,
2327 .get_port = inet_csk_get_port,
2328 .enter_memory_pressure = tcp_enter_memory_pressure,
2329 .stream_memory_free = tcp_stream_memory_free,
2330 .sockets_allocated = &tcp_sockets_allocated,
2331 .orphan_count = &tcp_orphan_count,
2332 .memory_allocated = &tcp_memory_allocated,
2333 .memory_pressure = &tcp_memory_pressure,
2334 .sysctl_mem = sysctl_tcp_mem,
2335 .sysctl_wmem = sysctl_tcp_wmem,
2336 .sysctl_rmem = sysctl_tcp_rmem,
2337 .max_header = MAX_TCP_HEADER,
2338 .obj_size = sizeof(struct tcp_sock),
2339 .slab_flags = SLAB_DESTROY_BY_RCU,
2340 .twsk_prot = &tcp_timewait_sock_ops,
2341 .rsk_prot = &tcp_request_sock_ops,
2342 .h.hashinfo = &tcp_hashinfo,
2343 .no_autobind = true,
2344 #ifdef CONFIG_COMPAT
2345 .compat_setsockopt = compat_tcp_setsockopt,
2346 .compat_getsockopt = compat_tcp_getsockopt,
2347 #endif
2348 .diag_destroy = tcp_abort,
2349 };
2350 EXPORT_SYMBOL(tcp_prot);
2351
2352 static void __net_exit tcp_sk_exit(struct net *net)
2353 {
2354 int cpu;
2355
2356 for_each_possible_cpu(cpu)
2357 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu));
2358 free_percpu(net->ipv4.tcp_sk);
2359 }
2360
2361 static int __net_init tcp_sk_init(struct net *net)
2362 {
2363 int res, cpu;
2364
2365 net->ipv4.tcp_sk = alloc_percpu(struct sock *);
2366 if (!net->ipv4.tcp_sk)
2367 return -ENOMEM;
2368
2369 for_each_possible_cpu(cpu) {
2370 struct sock *sk;
2371
2372 res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
2373 IPPROTO_TCP, net);
2374 if (res)
2375 goto fail;
2376 *per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk;
2377 }
2378
2379 net->ipv4.sysctl_tcp_ecn = 2;
2380 net->ipv4.sysctl_tcp_ecn_fallback = 1;
2381
2382 net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
2383 net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
2384 net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
2385
2386 net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME;
2387 net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES;
2388 net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL;
2389
2390 net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES;
2391 net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES;
2392
2393 net->ipv4.sysctl_tcp_syncookies = 0;
2394
2395 return 0;
2396 fail:
2397 tcp_sk_exit(net);
2398
2399 return res;
2400 }
2401
2402 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2403 {
2404 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2405 }
2406
2407 static struct pernet_operations __net_initdata tcp_sk_ops = {
2408 .init = tcp_sk_init,
2409 .exit = tcp_sk_exit,
2410 .exit_batch = tcp_sk_exit_batch,
2411 };
2412
2413 void __init tcp_v4_init(void)
2414 {
2415 inet_hashinfo_init(&tcp_hashinfo);
2416 if (register_pernet_subsys(&tcp_sk_ops))
2417 panic("Failed to create the TCP control socket.\n");
2418 }
Linux kernel - Deliverables
This page took 0.089424 seconds and 5 git commands to generate.